Domain Adaptation of Foundation LLMs for e-Commerce

TL;DR

This paper introduces e-Llama, domain-adapted large language models for e-commerce, achieved through extensive domain-specific pretraining, and evaluates their performance on specialized tasks while maintaining general capabilities.

Contribution

The paper presents the adaptation of Llama 3.1 models to e-commerce with 1 trillion tokens of data and introduces evaluation tasks to measure domain-specific performance.

Findings

Models adapt well without losing general performance

Careful training setup is crucial for effective domain adaptation

Merging base and adapted models offers performance trade-off control

Abstract

We present the e-Llama models: 8 billion and 70 billion parameter large language models that are adapted towards the e-commerce domain. These models are meant as foundation models with deep knowledge about e-commerce, that form a base for instruction- and fine-tuning. The e-Llama models are obtained by continuously pretraining the Llama 3.1 base models on 1 trillion tokens of domain-specific data. We discuss our approach and motivate our choice of hyperparameters with a series of ablation studies. To quantify how well the models have been adapted to the e-commerce domain, we define and implement a set of multilingual, e-commerce specific evaluation tasks. We show that, when carefully choosing the training setup, the Llama 3.1 models can be adapted towards the new domain without sacrificing significant performance on general domain tasks. We also explore the possibility of merging the adapted model and the base model for a better control of the performance trade-off between domains.